thresholding and 3dMean Group activations

AFNI Message Board
1

I’m trying to create ROIs for resting state analysis from each subject’s task fmri. I wanted to pick ROIs from the peak activation of the largest clusters after cluster correcting to p<0.05. The problem I ran into was how to restrict to regions that are considered “language.”

To solve this issue, I’ve made a mask from meta-analysis of the term “language.” Then I use 3dcalc to find where the meta-analysis mask and my stats file for each subject overlap. Now, my question is whether it is better to threshold to significant clusters before or after the mask. I’ve used 3dClust to make a cluster mask before intersecting with the meta-analysis mask, and the resulting clusters are pretty much the same except the peak voxel is different. Again, why is that? From a significance standpoint, what is the best approach here?

On a different note, if instead I wanted to create a group activation map for one task, how would I do that? I tried using 3dMean and it works up until about 15 subjects. As the number of subjects increases, the size of the clusters decreases until there is nothing (around 15 people). If I have closer to 40-50 subjects, how could I make a group map that would result in one or a few peak clusters?

Thank you!!

2

Your simple sounding question has only complicated answers.

[ol]
[li] How did you align the subjects? We recommend 3dQwarp (or its use in the @SSwarper or auto_warp.py scripts). Unless activation regions are large and strong, affine alignment does not do a good job.
[/li][li] 3dMean is not a statistical analysis program for groups. It is just a convenience program for averaging datasets together. In particular, if you are using 3dMean to average statistical maps from 3dDeconvolve or 3dREMLfit, you are mis-using the program.
[/li][li] Program 3dttest++ does simple group analyses, averaging beta maps across sets of subjects and carrying out voxel-wise t-tests. With the ‘-Clustsim’ option, it can also produce cluster-level significance maps from the voxel-wise t-maps.
[/li][/ol]
The above list is merely an outline of the procedures to carry out. We teach a class for a whole week on how to do this stuff. We have videos from the last time we taught this class, and are working on putting captions on them for distribution (required by Federal law). When they are available, we’ll post some links on this message board.